Consider the left shift operator $L:\mathbb{R}^\infty\to\mathbb{R}^\infty$ given by

$$ L(x_1,x_2,x_3,\ldots)=(x_2,x_3,\ldots). $$

Does the dual operator $L^\prime:(\mathbb{R}^\infty)^\prime\to(\mathbb{R}^\infty)^\prime$ have an eigenvector?

I want to say no via the following argument. An eigenvector $\varphi\in\left(\mathbb{R}^{\infty}\right)^{\prime}$ of $L^{\prime}$ would need to satisfy the eigenvalue equation $L^{\prime}\varphi=\alpha\varphi=\varphi\circ L$ for some $\alpha\in\mathbb{R}$. Applying this functional identity to an arbitrary element of $\mathbb{R}^{\infty}$ gives $$\alpha\varphi\left(x_{1},x_{2},x_{3},\ldots\right)=\varphi\left(x_{2},x_{3},x_{4},\ldots\right).$$

If we vary $x_{1}$, we see that the argument of the LHS of the above equation varies, but the argument of the RHS does not. Thus for our putative eigenvector $\varphi$ to satisfy the above functional identity, we require that $\varphi$ does not depend on the variable in the first slot of its argument. But this means that if we vary $x_{2}$, then the argument of the RHS of the above equation must remain unchanged, since $x_{2}$ is in the first slot of the argument of $\varphi$ in the RHS. But varying $x_{2}$ changes the second slot of the argument of the LHS, so we conclude that the above equation is only consistent if $\varphi$ does not depend on the variable in the second slot of its argument. Repeating this argument ad infinitum tells us that $\varphi$ must be independent of its argument entirely and hence a constant function $\varphi=c$. But the only constant functional that is also linear is $\varphi=0$, which cannot be an eigenvector since eigenvectors are defined to be nonzero. Hence $L^{\prime}$ has no eigenvectors.

After some thought however, I realized a flaw with this argument - namely, there could potentially exist a linear function which remains constant if I change a finite number of its arguments (what I proved) but nonetheless changes value if I simultaneously change an infinite number of its arguments. One such example that is non-linear is

$$ \varphi\left(x_{1},x_{2},\ldots\right)=\begin{cases} \lim_{n\to\infty}x_{n}, & \text{if this limit exists,}\\ 0, & \text{otherwise.} \end{cases} $$

My argument wouldn't rule out a function like this, but any such function I try to come up with ends up being non-linear.

Is it possible to tweak my argument by applying linearity so that I can rule out such pathological possibilities? Or does $L^{\prime}$ actually possess such a pathological, but linear, eigenvector (constructed perhaps via the axiom of choice)?

  • $\begingroup$ To be clear, by $\mathbb{R}^\infty$ you mean the vector space of all sequences, and $(\mathbb{R}^\infty)'$ is its algebraic dual? $\endgroup$ – Eric Wofsey Oct 1 '17 at 1:59
  • $\begingroup$ Yes, precisely. $\endgroup$ – sferics Oct 1 '17 at 2:45

Note that for any sequence $x=(x_1,x_2,x_3,\dots)$, there is another sequence $y=(y_1,y_2,y_3,\dots,)$ such that $x=\alpha y-Ly$. Namely, let $y_1$ be anything, and then recursively define $y_{n+1}=\alpha y_n-x_n$. If $\varphi$ is an eigenvector of $L'$ with eigenvalue $\alpha$ existed, then $\alpha\varphi(y)=\varphi(Ly)$ and so $0=\varphi(\alpha y-Ly)=\varphi(x)$. Since $x$ was arbitrary, this means $\varphi=0$.

To put it another way, the map $\alpha I-L$ is surjective, which implies its dual map $\alpha I-L'$ is injective, so $\alpha$ cannot be an eigenvalue of $L'$.

  • $\begingroup$ Simple, I like it! Thanks! $\endgroup$ – sferics Oct 1 '17 at 8:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.